- Home
- Guides
- Productivity Tools
- How to scrape website data into a spreadsheet with ChatGPT agent mode
How to scrape website data into a spreadsheet with ChatGPT agent mode
Give ChatGPT agent mode one clear prompt and it opens a web page, clicks the plus icon on each entry to reveal the details, and hands back a spreadsheet with the columns you asked for.
By Nur Islam- 3:21 video
- 5 steps
- 3.1K views
Jump to a step (5)
Directories and member lists often hide contact details behind a click on every entry. Copying the name, website, contact person, title, email and date for 49 states into Excel by hand takes hours and invites copy-paste mistakes.
ChatGPT agent mode runs a virtual browser. You describe the page, the click that reveals the data, and the exact fields you want, and it works through the page and returns a downloadable spreadsheet. You can watch its activity as it goes.
It is slower than a dedicated scraping tool because it interprets each entry, but it needs no setup or code.
What you need
- A ChatGPT account with agent mode available on your plan
- The URL of the page you want to collect data from
- A list of the exact fields you need
Step by step
0 of 5 done-
01
Work out what data you need and how to reveal it
Open the page and see how the data is shown. In the video each state has a plus icon that expands a card with the state name, website, president, contact person, contact person's job title, email address and state conference date.
-
02
Turn on agent mode
In ChatGPT, select the tools icon in the message box and choose Agent mode.
-
03
Write a specific prompt
Describe the task step by step. The prompt in the video tells the agent to:
- Visit the page URL.
- Click the plus icon to the right of each state name to expand the card.
- Collect the listed fields for every state.
- Return the results as a spreadsheet.
Name every field in the order you want the columns. If the data sits on child pages, say which links to open.
-
04
Run the agent and let it work
Paste the prompt and send it. The agent sets up a virtual desktop, opens the page and clicks each plus icon as instructed. You can follow its activity as it goes. For 49 states it took about 18 minutes.
-
05
Download and check the spreadsheet
When the agent finishes, download the file. In the video it has every requested column and 50 rows. Some cells are blank because the website did not list a job title for every contact. Spot-check a few rows against the website before you use the data.
Tips and common problems
- Be specific in the prompt. A vague prompt makes the agent spend extra time trying different ways to collect the data.
- Agent mode is slower than a dedicated scraper because it reads each entry before extracting it. For large bulk jobs a free scraping app is faster.
- Blank cells usually mean the source page had no value for that field, not that the agent missed it. Check a few against the site.
- If the details live on separate pages, tell the agent which links to follow and what to collect on each.
For bigger jobs, see a faster way to scrape website data with a free app.
Questions
Can ChatGPT scrape data from a website?
Yes, with agent mode. It opens the page in a virtual browser, follows your instructions such as clicking to expand entries, and returns the data as a spreadsheet you can download.
How long does ChatGPT agent take to scrape a page?
It depends on the page. In the video, collecting seven fields for 49 states behind expandable cards took about 18 minutes.
Do I need to write code to scrape with ChatGPT agent?
No. You describe the page, the clicks and the fields in plain English.
